Kotlin Insect ID Mobile App

Kotlin Insect ID is a mobile application built as part of the Trophées de la NSI challenge to recognize insects using a custom AI model trained us. The app is designed primarily for learning and experimentation, and is distributed under the MIT license.

Project goals

• Demonstrate how to integrate a locally running AI model (ONNX) into a modern Android app.
• Provide a practical NSI project that covers machine learning, mobile development, and UX.
• Offer a simple tool to identify insects from photos without requiring a constant internet connection.
• Easier to iterate than on than Kivy or Flet (python code, hard to develop a clean android app) for testing features and debugging.

Main features

• On‑device inference
  The app runs the insect classifier entirely on the device using our trained model.

• Image input from camera or gallery
  Users can:

  • Take a picture with the camera.
  • Pick an existing picture from the device storage.

• Insect recognition
  After selecting an image, the app:

  • Preprocesses the image for the model.
  • Runs inference locally.
  • Displays the predicted information (order, family, genre, specie) based on the model’s output.

• Online image fetching (reference images)
  Once an insect is recognized, the app can:

  • Build a query from the predicted labels.
  • Fetch example images from external image APIs (such as Pixabay or Unsplash, depending on configuration).
  • Display a small gallery/grid of related images so the user can visually compare and validate the prediction.
  • User can click on each of these images to open them in browser and later download

• History and debugging tools
  The app provides:

  • A simple history of past images.
  • Debug information to make it easier to understand model behaviour and troubleshoot issues during development.

Architecture overview

• Language & UI: Kotlin with Jetpack Compose for modern declarative UI.
• AI / Inference: ONNX Runtime to load and run the custom insect model entirely on device.
• Networking: HTTP client to call image search APIs and parse JSON responses.
• Image loading: Coil to efficiently display both local images and remote thumbnails in the UI.
• Storage: Simple local storage abstraction for saving and retrieving the last analyzed image and optional history.

Use cases

• NSI / educational project to:
  • Experiment with training image classification models for insects.
  • Learn how to deploy and use those models inside a native Android app.
  • Explore performance and UX trade‑offs between local processing and online resources.
  • Simply learn from insects

License

This project is released under the MIT License